MECÂNICA CLÁSSICA

UNIVERSIDADE FEDERAL DO RIO GRANDE DO SUL

INSTITUTO DE FÍSICA

PROFª DRª MARIA BEATRIZ DE LEONE GAY DUCATI

1
Action-angle variables for completely separable systems
Action-angle variables can also be introduced for certain types of motion of systems
with many degrees of freedom.

Complete separability means that the equations of canonical transformation have the form

𝜕𝑊𝑖 𝑞𝑖 ; 𝛼1 , . . . , 𝛼𝑛
𝑝𝑖 = , (10.98)
𝜕𝑞𝑖

which provides each 𝑝𝑖 as a function of the 𝑞𝑖 and the 𝑛 integration constants 𝛼𝑗 :

𝑝𝑖 = 𝑝𝑖 𝑞𝑖 ; 𝛼1 , . . . , 𝛼𝑛 . (10.99)

Here, Eq. (10.99) represents the orbit equation of the projection of the system point
on the (𝑝𝑖 , 𝑞𝑖 ) plane in phase space.

The action-angle variables lead to an evaluation of all the frequencies involved in

multiply periodic motion without requiring a complete solution of the motion. 2
Action-angle variables for completely separable systems (cont.)

In analogy to Eq. (10.82), the action variables 𝐽𝑖 are defined in terms of line
integrals over complete periods of the orbit in the (𝑞𝑖 , 𝑝𝑖 ) plane:

𝐽𝑖 = ර 𝑝𝑖 𝑑𝑞𝑖 . (10.100)

If one of the separation coordinates is cyclic, its conjugate momentum is constant.

The corresponding orbit in the 𝑞𝑖 , 𝑝𝑖 plane of phase space is then a horizontal

straight line, which would not appear to be in the nature of a periodic motion.

Accordingly, the integral in the definition of the action variable corresponding to

a cyclic angle coordinate is to be evaluated from 0 to 2𝜋, and hence

𝐽𝑖 = 2𝜋𝑝𝑖 (10.101)

for all cyclic variables.

3
Action-angle variables for completely separable systems (cont.)

By Eq. (10.98), 𝐽𝑖 can also be written as:

𝜕𝑊𝑖 𝑞𝑖 ; 𝛼1 , . . . , 𝛼𝑛
𝐽=ර 𝑑𝑞𝑖 . (10.102)
𝜕𝑞𝑖

Expressing the 𝛼1 's as functions of the action variables, the characteristic

function 𝑊 can be written in the form

𝑊 = 𝑊 𝑞1 , … , 𝑞𝑛 ; 𝐽1 , … , 𝐽𝑛 = ෍ 𝑊𝑗 𝑞𝑗 ; 𝐽1 , . . . , 𝐽𝑛 ,
𝑗

while the Hamiltonian appears as a function of the 𝐽𝑖 's only:

𝐻 = 𝛼1 = 𝐻 𝐽1 , . . . , 𝐽𝑛 . (10.103)

4
Action-angle variables for completely separable systems (cont.)

As in the system of one degree of freedom, we can define conjugate angle

variables 𝑤𝑖 by the equations of transformation that here appear as:
𝑛
𝜕𝑊 𝜕𝑊𝑗 𝑞𝑗 ; 𝐽1 , . . . , 𝐽𝑛
𝑤𝑖 = =෍ . (10.104)
𝜕𝐽𝑖 𝜕𝐽𝑖
𝑗=1
Note in general 𝑤𝑖 could be a function of several or all of the 𝑞𝑖 ; that is, 𝑤𝑖 =
𝑤𝑖 𝑞𝑖 , … , 𝑞𝑛 ; 𝐽𝑖 , … , 𝐽𝑛 .

The 𝑤𝑖 's satisfy equations of motion given by

𝜕𝐻 𝐽1 , . . . , 𝐽𝑛
𝑤ሶ 𝑖 = = 𝑣𝑖 𝐽1 , . . . , 𝐽𝑛 . (10.105)
𝜕𝐽𝑖
Because the 𝑣𝑖 's are constants, functions of the action variables only, the angle variables are
all linear functions of time
𝑤𝑖 = 𝑣𝑖 𝑡 + 𝛽𝑖 . (10.106)
5
Action-angle variables for completely separable systems (cont.)

In effect, we are dealing with analogues of the virtual displacements, and

accordingly the infinitesimal change in 𝑤𝑖 as the 𝑞𝑗 's are changed infinitesimally will be
denoted by 𝛿𝑤𝑖 and is given by

𝜕𝑤𝑖 𝜕²𝑊
𝛿𝑤𝑖 = ෍ 𝑑𝑞𝑗 = ෍ 𝑑𝑞𝑗 ,
𝜕𝑞𝑗 𝜕𝐽𝑖 𝜕𝑞𝑗
𝑗

where use has been made of Eq. (10.104). The derivative with respect to 𝑞𝑖 vanishes
except for the 𝑊𝑗 constituent of 𝑊, so that by Eq. (10.98) 𝛿𝑤𝑖 reduces to

𝜕
𝛿𝑤𝑖 = ෍ 𝑝𝑗 𝑞𝑗 , 𝐽 𝑑𝑞𝑗 . (10.107)
𝜕𝐽𝑖
𝑗

6
Action-angle variables for completely separable systems (cont.)

Equation (10.107) represents 𝛿𝑤𝑖 as the sum of independent contributions each

involving the 𝑞𝑗 motion. The total change in 𝑤𝑖 as a result of the specified maneuver is
therefore
𝜕
∆𝑤𝑖 = ෍ ර 𝑝𝑗 𝑞𝑗 , 𝐽 𝑑𝑞𝑗 . (10.108)
𝜕𝐽𝑖
𝑗 𝑚𝑗

Below each integral sign, the symbol 𝑚𝑗 indicates the integration is over 𝑚𝑗 cycles of 𝑞𝑗 .

Each of the integrals is, by the definition of the action variables, exactly 𝑚𝑗 𝐽𝑗 .
Since the 𝐽 's are independent, it follows that
∆𝑤𝑖 = 𝑚𝑖 . (10.109)
If the sets of 𝑤's and 𝑚's are treated as vectors 𝐰 and 𝐦, Eq. (10.109) can be written as
∆𝐰 = 𝐦. (10.109’)
7
Action-angle variables for completely separable systems (cont.)

Suppose that the separable motions are all of the libration type so that each 𝑞𝑗 ,
as well as 𝑝𝑗 , returns to its initial value on completion of a complete cycle.

Such a multiply periodic function can always be represented by a multiple

Fourier expansion, which for 𝑞𝑘 , say, would appear as
∞ ∞ ∞
(𝑘)
𝑞𝑘 = ෍ ෍ , … , ෍ 𝑎𝑗1 ,…,𝑗𝑛 ∙ 𝑒 2𝜋𝑖 𝑗1 𝑤1 +𝑗2 𝑤2 +𝑗3 𝑤3 +⋯+𝑗𝑛 𝑤𝑛
, (libration) (10.110)
𝑗1 =−∞ 𝑗2 =−∞ 𝑗𝑛 =−∞

where the 𝑗's are 𝑛 integer indices running from −∞ to ∞. By treating the set of 𝑗's also
as a vector in the same 𝑛-dimensional space with 𝐰, the expansion can be written more
compactly as
(𝑘)
𝑞𝑘 = ෍ 𝑎𝐣 𝑎2𝜋𝑖𝐣∙𝐰 , libration . (10.110’)
𝐣
8
Action-angle variables for completely separable systems (cont.)
If we similarly write Eq. (10.109') as a vector equation,
𝐰 = 𝐯𝑡 + 𝛃, (10.106’)
then the time dependence of 𝑞𝑘 appears in the form
𝑘
𝑞𝑘 𝑡 = ෍ 𝑎𝐣 𝑒 2𝜋𝐣∙ 𝐯𝑡+𝛃
, libration . (10.111)
𝐣

Note that in general 𝑞𝑘 𝑡 is not a periodic function of 𝑡. Unless the various 𝜈𝑖 ’s are
commensurate (that is, rational multiples of each other), 𝑞𝑘 will not repeat its values at
regular intervals of time.

Considered as a function of 𝑡, is designated as a quasi-periodic function. That the

𝑘
coefficients 𝑎𝑗 can be found by the standard procedure for Fourier coefficients; they are
given by the multiple integral over the unit cell in 𝐰 space:
1 1
(𝑘)
𝑎𝐣 = න , … , න 𝑞𝑘 (𝐰)𝑒 −2𝜋𝑖𝐣∙𝐰 (𝑑𝐰) , (10.112)
0 0

Here (𝑑𝐰) stands for the volume element in the 𝑛-dimensional space of the 𝑤𝑖 's. 9
Action-angle variables for completely separable systems (cont.)

When the motion is in the nature of a rotation, then in a complete cycle of the
separated variable pair (𝑞𝑘 , 𝑝𝑘 ) the coordinate 𝑞𝑘 does not return to its original value,
but instead increases by the value of its period 𝑞0𝑘 . However, during the cycle we have
seen that 𝑤𝑘 increases by unity. Hence, the function 𝑞𝑘 − 𝑤𝑘 𝑞0𝑘 does return to its initial
value and, like the librational coordinates, is a multiply periodic function of all the 𝑤's
with unit periods. We can therefore expand the function in a multiple Fourier series
analogous to Eq. (10.110)

𝑘
𝑞𝑘 − 𝑤𝑘 𝑞0𝑘 = ෍ 𝑎𝐣 𝑒 2𝜋𝑖𝐣∙𝒘 , rotation (10.113)
𝐣
or

(𝑘)
𝑞𝑘 = 𝑞0𝑘 𝑣𝑘 𝑡 + 𝛽𝑘 + ෍ 𝑎𝐣 𝑒 2𝜋𝑖𝐣∙ 𝐯𝑡+𝛃
, rotation . (10.114)
𝐣
10
Action-angle variables for completely separable systems (cont.)

The separable momentum coordinates, 𝑝𝑘 , are by the nature of the assumed

motion also multiply periodic functions of the 𝑤's and can be expanded in a multiple
Fourier series similar to Eq. (10.110). It follows then that any function of the several
variable pairs (𝑞𝑘 , 𝑝𝑘 ) will also be multiply periodic functions of the 𝑤's and can be
written in the form

𝑓 𝑞, 𝑝 = ෍ 𝑏𝐣 𝑒 2𝜋𝑖𝐣∙𝐰 = ෍ 𝑏𝐣 𝑒 2𝜋𝑖𝐣∙ 𝐯𝑡+𝛃 . (10.115)

𝐣 𝐣

While Eqs. (10.110) and (10.111) represent the most general type of motion
consistent with the assumed nature of the problem, not all systems will exhibit this full
generality. In particular, for most problems simple enough to be used as illustrations of
the application of action-angle variables, Eq. (10.104) simplifies to
𝜕𝑤𝑖
𝑤𝑖 = 𝑞𝑖 ; 𝐽1 , . . . , 𝐽𝑛 , (10.116)
𝜕𝐽𝑖
11
Action-angle variables for completely separable systems (cont.)

and each separation coordinate 𝑞𝑖 is a function only of its corresponding 𝑤𝑘 . When this
happens, 𝑞𝑘 is then a periodic function of 𝑤𝑘 (and therefore of time), and the multiple
Fourier series reduces to a single Fourier series:
(𝑘) (𝑘)
𝑞𝑘 = ෍ 𝑎𝑗 𝑒 2𝜋𝑖𝑗𝑤𝑘 = ෍ 𝑎𝑗 𝑒 2𝜋𝑖𝑗(𝑣𝑘 𝑡+𝛽𝑘 ) . (10.117)
𝑗 𝑗

The motion of a two-dimensional anisotropic harmonic oscillator provides a

convenient and familiar example of these considerations.

Suppose that in a particular set of Cartesian coordinates the Hamiltonian is

given by
1
𝐻= 𝑝𝑥2 + 4𝜋 2 𝑚2 𝑣𝑥2 𝑥 2 + 𝑝𝑦2 + 4𝜋 2 𝑚2 𝑣𝑦2 𝑦 2 .
2𝑚

12
Action-angle variables for completely separable systems (cont.)

Each Cartesian coordinate will exhibit simple harmonic motion with

frequencies 𝜈𝑥 and 𝜈𝑦 , respectively. Thus, the solutions for 𝑥 and 𝑦 are particularly
simple forms of the single Fourier expansions of Eq. (10.117). Suppose now that the
coordinates are rotated 45° about the 𝑧 axis; the components of the motion along the
new 𝑥 ′ , 𝑦′ axes will be

1
𝑥′ = 𝑥0 𝑐𝑜𝑠2𝜋 𝜈𝑥 𝑡 + 𝛽𝑥 + 𝑦0 𝑐𝑜𝑠2𝜋 𝜈𝑦 𝑡 + 𝛽𝑦
2
(10.118)
1
𝑦′ = 𝑦0 𝑐𝑜𝑠2𝜋 𝜈𝑦 𝑡 + 𝛽𝑦 − 𝑥0 𝑐𝑜𝑠2𝜋 𝜈𝑥 𝑡 + 𝛽𝑥
2

13
Action-angle variables for completely separable systems (cont.)

If 𝜈𝑥 /𝜈𝑦 is a rational number, these two expressions will be commensurate,

corresponding to closed Lissajous figures of the type shown in Fig. 10.4. But if 𝜈𝑥 and 𝜈𝑦
are incommensurable, the Lissajous figure never exactly retraces its steps and Eqs.
(10.118) provide simple examples of multiply periodic series expansions of the form
(10.117).

14
Action-angle variables for completely separable systems (cont.)

Suppose that the time interval 𝑇 contains 𝑚 complete cycles of 𝑞𝑘 plus a fraction
of a cycle.

In general, the times required for each successive cycle will be different, since
𝑞𝑘 will not be a periodic function of 𝑡.

Based in number theory, that as 𝑇 increases indefinitely,

𝑚
Lim = 𝜈𝑘 .
𝑡→∞ 𝑇
(10.119)

A multiply periodic function can always be formed from the generating function
𝑊. The defining equation for 𝐽𝑖 , Eq. (10.102) states that when 𝑞𝑖 goes through a
complete cycle; that is, when 𝑤𝑖 ; changes by unity, the characteristic function increases
by 𝐽𝑖 . It follows that the function

𝑊 = 𝑊 ′ − ෍ 𝑤𝑘 𝐽𝑘 . (10.120)
15
𝑘
Action-angle variables for completely separable systems (cont.)

Equation (10.120) therefore represents a multiply periodic function that can be

expanded in terms of the 𝑤𝑖 (or of the frequencies 𝜈𝑖 ) by a series of the form of Eq. (
10.115). Since the transformation equations for the angle variables are
𝜕𝑊
𝑤𝑘 =
𝜕𝐽𝑘

it will be recognized that Eq. (10.120) defines a Legendre transformation from the 𝑞, 𝐽
basis to the 𝑞, 𝑤 basis.

The formal condition for the commensurability of two frequencies 𝜈𝑖 and 𝜈𝑗 is

that they satisfy the relation 𝑗𝑖 𝜈𝑖 = 𝑗𝑗 𝜈𝑗 (no sum) where 𝑗𝑖 and 𝑗𝑗 are nonzero positive
integers. All pairs of frequencies must satisfy relations of the form

𝑗𝑖 𝜈𝑖 = 𝑗𝑘 𝜈𝑘 , no sum . (10.121)
16
Action-angle variables for completely separable systems (cont.)

When we can express any 𝜈𝑖 as a rational fraction of any of the other frequencies,
the system is said to be completely commensurate.

If only 𝑚 + 1 of the 𝑛 frequencies satisfy Eq. (10.121), the system is said to be 𝑚-fold
commensurate.

For example, consider the set of seven frequencies 𝜈1 = 3 MHz, 𝜈2 = 5 MHz, 𝜈3 =

7 MHz , 𝜈4 = 2 2 MHz , 𝜈5 = 3 2 MHz , 𝜈6 = 3 MHz , 𝜈7 = 7 MHz . The first three 𝜈1 , 𝜈2 , and
𝜈3 are triply commensurate, the next two 𝜈4 and 𝜈5 are doubly commensurate.

In degenerate systems, the fundamental frequencies are no longer independent,

and the periodic motion of the system can be described by less than the full complement of 𝑛
frequencies. Indeed, the 𝑚 conditions of degeneracy can be used to reduce the number of
frequencies to n − 𝑚 + 1. The 𝑚 degeneracy conditions may be written where 𝑗𝑘𝑖 are positive
or negative integers 𝑛

෍ 𝑗𝑘𝑖 𝜈𝑖 = 0, 𝑘 = 1, … , 𝑚. (10.122)
17
𝑖=1
Action-angle variables for completely separable systems (cont.)

Consider now a point transformation from (𝑤, 𝐽) to (𝑤 ′ , 𝐽′ ) defined by the

generating function (cf. Eq. (9.26) where the summation convention is used):
𝑚 𝑛 𝑛

𝐹2 = ෍ ෍ 𝐽𝑘′ 𝑗𝑘𝑖 𝑤𝑖 + ෍ 𝐽𝑘′ 𝑤𝑘 . (10.123)

𝑘=1 𝑖=1 𝑘=𝑚+1

The transformed coordinates are

𝑛

𝑤𝑘′ = ෍ 𝑗𝑘𝑖 , 𝑘 = 1, … , 𝑚
(10.124)
𝑖=1
= 𝑤𝑘 , 𝑘 = 𝑚 + 1, … , 𝑚.
Correspondingly, the new frequencies are
𝑛

𝜈𝑘′ = 𝑤ሶ 𝑘′ = ෍ 𝑗𝑘𝑖 𝜈𝑖 = 0, 𝑘 = 1, … , 𝑚
(10.125)
𝑖=1
= 𝜈𝑘 , 𝑘 = 𝑚 + 1, … , 𝑚. 18
Action-angle variables for completely separable systems (cont.)

The corresponding constant action variables are given as the solution of the 𝑛
equations of transformation
𝑚 𝑛

𝐽𝑖 = ෍ 𝐽𝑘′ 𝑗𝑘𝑖 + ෍ 𝐽𝑘′ 𝛿𝑘𝑖 (10.126)

𝑘=1 𝑘=𝑚+1

These are of course also present in the original Fourier series in terms of the 𝜈's,
Eq. (10.110), occurring whenever the indices 𝑗𝑖 are such that degeneracy conditions are
satisfied. Since
𝜕𝐻
𝜈𝑖′ = ′,
𝜕𝐽𝑖

the Hamiltonian must be independent of the action variables 𝐽𝑖′ whose corresponding
frequencies vanish. In a completely degenerate system, the Hamiltonian can therefore
be made to depend upon only one of the action variables.
19
Action-angle variables for completely separable systems (cont.)

Note that Hamilton's characteristic function 𝑊 also serves as the generating

function for the transformation from the (𝑞, 𝑝) set to the (𝑤 ′ , 𝐽′ ) set.

The 𝐽′ quantities are 𝑛 independent constants, the original constants of

integration may be expressed in terms of the 𝐽′ set, and 𝑊 given as 𝑊 (𝑞, 𝐽′ ).

It is a generating function to a new set of canonical variables for which the 𝐽′

quantities are the canonical momenta. But by virtue of the point transformation
generated by the 𝐹2 of Eq. (10.123), we know that 𝑤 ′ is conjugate to 𝐽′ .

It follows that the new coordinates generated by 𝑊 (𝑞, 𝐽′ ) must be the angle
variable 𝑤 ′ set, with equations of transformation given by

𝜕𝑊
𝑤𝑖′ = ′. (10.127)
𝜕𝐽𝑖
20
The Kepler problem in action-angle variables
In terms of spherical polar coordinates, the Kepler problem becomes a special
case of the general treatment given above in Section 10.5 for central force motion in
space. Equations (10.70) through (10.77) can be taken over here immediately, replacing
𝑉 𝑟 wherever it occurs by its specific form
𝑘
𝑉 𝑟 =− , (10.128)
𝑟
The potential 𝑉 𝑟 depends only upon one of the three coordinates, and the Hamilton-
Jacobi equation is completely separable in spherical polar coordinates.

Confining the discussion to the bound case, 𝐸 > 0.

The motion in each of the coordinates will be periodic-libration in 𝑟 and 𝜃, and

rotation in 𝜙. From Eq. (10.72), it follows that
𝜕𝑊
𝐽𝜙 = ර 𝑑𝜙 = ර 𝛼𝜙 𝑑𝜙 . (10.129a)
𝜕𝜙 21
The Kepler problem in action-angle variables (cont.)

Similarly, on the basis of Eq. (10.74), 𝐽𝜃 is given by

2
𝜕𝑊 𝛼 𝜙
𝐽𝜃 = ර 𝑑𝜃 = ර 𝛼𝜃2 − 𝑑𝜃 . (10.129b)
𝜕𝜃 2
𝑠𝑖𝑛 𝜃

Finally the integral for 𝐽𝑟 from Eq. (10.75), is

𝜕𝑊 2𝑚𝑘 𝛼𝜃2
𝐽𝑟 = ර 𝑑𝑟 = ර 2𝑚𝐸 + − 2 𝑑𝑟 . (10.129c)
𝜕𝑟 𝑟 𝑟

The first integral is trivial; 𝜙 goes through 2𝜋 radians in a complete revolution

and therefore

𝐽𝜙 = 2𝜋𝛼𝜙 = 2𝜋𝑝𝜙 . (10.130)

22
The Kepler problem in action-angle variables (cont.)

Integration of Eq. (10.129b) can be performed in various ways. If the polar angle
of the total angular momentum vector is denoted by i, so that
𝛼𝜙
cos 𝑖 = , (10.131)
𝛼𝜃
then Eq. (10.129b) can also be written as

𝐽𝜃 = 𝛼𝜃 ර 1 − 𝑐𝑜𝑠 2 𝑖 𝑐𝑠𝑐 2 𝜃𝑑𝜃 . (10.132)

The complete circuital path of integration is for 𝜃 going from a limit −𝜃0 to +𝜃0 and
back again, where sin 𝜃0 = cos 𝑖, or 𝜃0 = 𝜋/2 − 𝑖. Hence, the circuital integral can be
written as 4 times the integral over from 0 to 𝜃0 , or after some manipulation,
𝜃0
𝐽𝜃 = 4𝛼𝜃 න csc 𝜃 𝑠𝑖𝑛2 𝑖 − 𝑐𝑜𝑠 2 𝜃 𝑑𝜃 .
0
23
The Kepler problem in action-angle variables (cont.)

The substitution

cos 𝜃 = sin 𝑖 sin 𝜓

transforms the integral to

𝜋/2
𝑐𝑜𝑠 2 𝜓 𝑑𝜓
𝐽𝜃 = 4𝛼𝜃 𝑠𝑖𝑛2 𝑖 න 2 2
. (10.133)
0 1 − 𝑠𝑖𝑛 𝑖 𝑠𝑖𝑛 𝜓

Finally, with the substitution

𝑢 = tan 𝜓 ,

24
The Kepler problem in action-angle variables (cont.)

the integral becomes

∞
2
𝑑𝑢
𝐽𝜃 = 4𝛼𝜃 𝑠𝑖𝑛 𝑖 න
0 1 + 𝑢2 1 + 𝑢2 𝑐𝑜𝑠 2 𝑖
(10.134)
∞
1 𝑐𝑜𝑠 2 𝑖
= 4𝛼𝜃 න 𝑑𝑢 2
− 2 𝑐𝑜𝑠 2 𝑖
.
0 1 + 𝑢 1 + 𝑢

This last form involves only well-known integrals, and the final result is

𝐽𝜃 = 2𝜋𝛼𝜃 1 − cos 𝑖 = 2𝜋 𝛼𝜃 − 𝛼𝜙 . (10.135)

The last integral (Eq. (10.129c)), for 𝐽𝑟 , can now be written as

2
2𝑚𝑘 𝐽𝜃 + 𝐽𝜙
𝐽𝑟 = ර 2𝑚𝐸 + − 𝑑𝑟 . (10.136)
𝑟 4𝜋 2 𝑟 2
25
The Kepler problem in action-angle variables (cont.)

The integral involved in Eq. (10.136) can be evaluated by elementary means,

but the integration is most elegantly and quickly performed using the complex contour
integration method of residues.

The integration involves both branches of a double-valued function, with 𝑟1 and

𝑟2 as the branch points. Consequently, the complex plane can be represented as one of
the sheets of a Riemann surface, slit along the real axis from 𝑟1 to 𝑟2 (as indicated in
Fig. 10.5).

26
The Kepler problem in action-angle variables (cont.)

The path of integration encloses the line between the branch points 𝑟1 and 𝑟2 ,
then the method of residues cannot be applied directly.

Considering the path as enclosing all the rest of the complex plane, the
direction of integration now being in the reverse (clockwise) direction. Now, the sign in
front of the square root in the integrand must be negative for the region along the real
axis below 𝑟1 , as can be seen by examining the behavior of the function in the
neighborhood of 𝑟1 .

If the integrand is represented as

2𝐵 𝐶
− 𝐴+ − 2,
𝑟 𝑟
the residue at the origin is

𝑅0 = − −𝐶.
27
The Kepler problem in action-angle variables (cont.)

Above 𝑟2 , the sign of the square root on the real axis is found to be positive, and
the residue is obtained by the standard technique of changing the variable of
integration to 𝑧 = 𝑟 −1 :
1
− ර 2 𝐴 + 2𝐵𝑧 − 𝐶𝑧 2 𝑑𝑧 . (10.137)
𝑧

Expansion about 𝑧 = 0 now furnishes the residue

𝐵
𝑅∞ = − .
𝐴
The total integral is −2𝜋𝑖 times the sum of the residues:

𝐵
𝐽𝑟 = 2𝜋𝑖 −𝐶 + , (10.138)
𝐴
28
The Kepler problem in action-angle variables (cont.)

or, upon substituting the coefficients 𝐴, 𝐵, and 𝐶:

2𝑚
𝐽𝑟 = − 𝐽𝜃 + 𝐽𝜙 + 𝜋𝑘 (10.139)
−𝐸

Equation (10.139) supplies the functional dependence of 𝐻 upon the action

variables; for solving for 𝐸, we have

2𝜋 2 𝑚𝑘 2
𝐻≡𝐸=− 2 (10.140)
𝐽𝑟 + 𝐽𝜃 + 𝐽𝜙

Note that all three of the action variables appear only in the form 𝐽𝑟 + 𝐽𝜃 + 𝐽𝜙 . Hence, all
of the frequencies are equal; the motion is completely degenerate.

29
The Kepler problem in action-angle variables (cont.)

If the central force contained an 𝑟 −3 term, such as is provided by first-order

relativistic corrections, then the orbit is in the form of a precessing ellipse. One of the
degeneracies will be removed in this case, but the motion is still singly degenerate, since
𝜈𝜃 = 𝜈𝜙 for all central forces.

The one frequency for the motion here is given by

𝜕𝐻 𝜕𝐻 𝜕𝐻 4𝜋 2 𝑚𝑘 2
𝜈= = = =− 3. (10.141)
𝜕𝐽𝑟 𝜕𝐽𝜃 𝜕𝐽𝜙 𝐽𝑟 + 𝐽𝜃 + 𝐽𝜙

If we evaluate the sum of the 𝐽 's in terms of the energy from Eq. (10.140) the period of the
orbit is
𝑚
𝜏 = 𝜋𝑘 . (10.142)
−2𝐸 3

This formula for the period agrees with Kepler's third law, Eq. (3.71), if it is remembered
that the semimajor axis a is equal to −𝑘/2𝐸. 30
The Kepler problem in action-angle variables (cont.)

The degenerate frequencies may be eliminated by canonical transformation to a

new set of action-angle variables, following the procedure outlined in the previous
section. Expressing the degeneracy conditions as

𝑣𝜙 − 𝑣𝜃 = 0, 𝑣𝜃 − 𝑣𝑟 = 0,

the appropriate generating function is

𝐹 = 𝑤𝜙 − 𝑤𝜃 𝐽1 + 𝑤𝜃 − 𝑤𝑟 𝐽2 + 𝑤𝑟 𝐽3 . (10.143)

The new angle variables are

𝑤1 = 𝑤𝜙 − 𝑤𝜃 , 𝑤2 = 𝑤𝜃 − 𝑤𝑟 , 𝑤3 = 𝑤𝑟 , (10.144)

and, as planned, two of the new frequencies, 𝜈1 and 𝜈2 , are zero.

31
The Kepler problem in action-angle variables (cont.)

We can obtain the new action variables from the transformation equations

𝐽𝜙 = 𝐽1 , 𝐽𝜃 = 𝐽2 − 𝐽1 , 𝐽𝑟 = 𝐽3 − 𝐽2 , .

which yields the relations

𝐽1 = 𝐽𝜙 , 𝐽2 = 𝐽𝜙 + 𝐽𝜃 , 𝐽3 = 𝐽𝜙 + 𝐽𝜃 + 𝐽𝑟 (10.145)

In terms of these transformed variables the Hamiltonian appears as

2𝜋 2 𝑚𝑘 2
𝐻=− (10.146)
𝐽32
a form involving only that action variable for which the corresponding frequency is
different from zero.
32
The Kepler problem in action-angle variables (cont.)

But the motion for the bound Kepler problem is a particular closed orbit in a
plane, then the integrals for 𝐽𝜙 and 𝐽𝑟 can be evaluated very quickly and simply. For the
𝐽𝜃 integral, we can apply the following procedure. It will be recalled that when the
defining equations for the generalized coordinates do not involve time explicitly, then
(cf. Eq. (8.20) and the material following (8.20))

𝑝𝑖 𝑞ሶ 𝑖 = 2𝐿2 𝑞ሶ 𝑖 𝑞ሶ 𝑖 = 2𝑇.

We can express the kinetic energy 𝑇 either in spherical polar coordinates or in the plane
polar coordinates (𝑟, 𝜓). It follows, then, that

2𝑇 = 𝑝𝑟 𝑟ሶ + 𝑝𝜃 𝜃ሶ + 𝑝𝜙 𝜙ሶ = 𝑝𝑟 𝑟ሶ + 𝑝𝜓,ሶ (10.147)

where 𝑝(≡ 𝑙) is the magnitude of the total angular momentum.

33
The Kepler problem in action-angle variables (cont.)

Hence, the definition for 𝐽𝜃 can also be written as

𝐽𝜃 ≡ ර 𝑝𝜃 𝑑𝜃 = ර 𝑝 𝑑𝜓 − ර 𝑝𝜙 𝑑𝜙 (10.148)

Because the frequencies for 𝜃 and 𝜙 are equal, both 𝜙 and 𝜓 vary by 2𝜋 as 𝜃 goes
through a complete cycle of libration, and the integrals defining 𝐽𝜃 reduce to

𝐽𝜃 = 2𝜋 𝑝 − 𝑝𝜙 = 2𝜋 𝛼𝜃 − 𝛼𝜙 .

The virial theorem for the bound orbits in the Kepler problem says that (cf. Eq.
(3.30))
𝑉ത = −2𝑇,
ത

where the bar denotes an average over a single complete period of the motion.

34
The Kepler problem in action-angle variables (cont.)

It follows that

𝐻 ≡ 𝐸 = 𝑇ത + 𝑉ത = −𝑇.
ത (10.149)

Integrating Eq. (10.147) with respect to time over a complete period of motion we have

2𝑇ത
= 𝐽𝑟 + 𝐽𝜃 + 𝐽𝜙 = 𝐽3 , (10.150)
𝜈3
where 𝜈3 is the frequency of the motion, that is, the reciprocal of the period. Combining
Eqs. (10.149) and (10.150) leads to the relation
2 𝑣3 1 𝑑𝐻
− = = , (10.151)
𝐽3 𝐻 𝐻 𝑑𝐽3

where use has been made of Eq. (10.105). Equation (10.151) is in effect a differential
equation for the functional behavior of 𝐻 on 𝐽3 .
35
The Kepler problem in action-angle variables (cont.)

Integration of the equation immediately leads to the solution:

𝐷
𝐻 = 2, (10.152)
𝐽3

where 𝐷 is a constant that depend only upon 𝑚 and 𝑘. We can evaluate 𝐷 by considering
the elementary case of a circular orbit, of radius 𝑟0 , for which 𝐽𝑟 = 0 and 𝐽3 = 2𝜋𝑝. The
total energy is here
𝑘
𝐻=− (10.153)
2𝑟0

(as can most immediately been seen from the virial theorem). Further, the condition for
circularity, Eq. (3.40), can be written for the inverse-square force law as

𝑘 𝑝2 𝐽3
2 = 3 = 2 3. (10.153)
𝑟0 𝑚𝑟0 4𝜋 𝑚𝑟0
36
The Kepler problem in action-angle variables (cont.)

Eliminating 𝑟0 between Eqs. (10.153) and (10.154) leads to

2𝜋 2 𝑚𝑘 2
𝐻=− (10.155)
𝐽32

This result has been derived only for circular orbits. But Eq. (10.152) says it
must also be correct for all bound orbits of the Kepler problem, and indeed it is identical
with Eq. (10.146). Thus, if the existence of a single period for all coordinates is taken as
known beforehand, it is possible to obtain 𝐻(𝐽) without direct evaluation of the circuital
integrals.

37
The Kepler problem in action-angle variables (cont.)

One usual set of astronomical elements therefore consists of the six constants

𝑖, Ω, 𝑎, 𝑒, 𝜔, 𝑇,

where the last one, 𝑇, is the

time of passage through the
periapsis point. Of the
remaining five, the first two
defines the orientation of the
orbital plane in space, while 𝑎,
𝑒, and 𝜔 directly specify the
scale, shape, and orientation of
the elliptic orbit, respectively
(Figure 10.6).

38
The Kepler problem in action-angle variables (cont.)

The action-angle variable treatment of the Kepler problem also leads to five
algebraic constants of the motion.

Three of them is the three constant action variables, 𝐽1 , 𝐽2 , and 𝐽3 . The

remaining two are the angle variables 𝑤1 and 𝑤2 , which are constants, because their
corresponding frequencies are zero. 𝐽1 , 𝐽2 , 𝐽3 , 𝑤1 and 𝑤2 in terms of the classical orbital
elements 𝑖, Ω, 𝑎, 𝑒, 𝜔, and vice versa. Some of these interrelations are immediately
obvious.

From Eqs. (10.145) and (10.135) it follows that

𝐽2 = 2𝜋𝛼𝜃 ≡ 2𝜋𝑙, (10.156)

and hence, by Eq. (10.131),

𝐽1
= cos 𝑖 . (10.157)
𝐽2 39
The Kepler problem in action-angle variables (cont.)

As is well known, the semimajor axis a is a function only of the total energy 𝐸
(cf. Eq. (3.61)) and therefore, by Eq. (10.146), 𝑎 is given directly in terms of 𝐽3 :
𝑘 𝐽32
𝑎=− = 2 . (10.158)
2𝐸 4𝜋 𝑚𝑘
In terms of 𝐽2 , Eq. (3.62) for the eccentricities can be written as

𝐽22
𝑒= 1− 2 ,
4𝜋 𝑚𝑘𝑎

or
2
𝐽2
𝑒= 1− . (10.159)
𝐽3
40
The Kepler problem in action-angle variables (cont.)

The equation of transformation defining 𝑤1 is, by Eq. (10.127),

𝜕𝑊
𝑤1 = .
𝜕𝐽1
It can be seen from the separated form of 𝑊, Eq. (10.71), that 𝑊 can be written as the
sum of indefinite integrals:

𝑊 = න 𝑝𝜙 𝑑𝜙 + න 𝑝𝜃 𝑑𝜃 + න 𝑝𝑟 𝑑𝑟 . (10.160)

As we have seen from the discussion on 𝐽𝑟 , the radial momentum 𝑝𝑟 does not
involve 𝐽1 , but only 𝐽3 (through 𝐸) and the combination 𝐽𝜃 + 𝐽𝜙 = 𝐽2 . Only the first two
integrals are therefore involved in the derivative with respect to 𝐽1 . By Eq. (10.130),
𝐽1
𝑝𝜙 = 𝛼𝜙 = . (10.161)
2𝜋
41
The Kepler problem in action-angle variables (cont.)

And by Eq. (10.74), with the help of Eqs. (10.156) and (10.161),

2 2
𝛼𝜙 1 𝐽
𝑝𝜃 = ± 𝛼𝜃2 − = ± 𝐽 2
−
1 (10.162)
𝑠𝑖𝑛2 𝜃 2𝜋 2 sin2 𝜃

It turns out that in order to relate 𝑤1 to the ascending node, it is necessary to choose
the negative sign of the square root. The angular variable 𝑤1 is therefore determined by

𝜙 𝐽1 𝑑𝜃
𝑤1 = + න ,
2𝜋 2𝜋 sin2 𝜃 𝐽2 − 𝐽2 csc 2 𝜃
2 1

or

𝑑𝜃 cot 𝑖 csc 2 𝜃 𝑑𝜃
2𝜋𝑤1 = 𝜙 + cos 𝑖 න =𝜙+න .
sin2 𝜃 1 − cos 2 𝑖 𝑐𝑠𝑐 2 𝜃 1 − cot 2 𝑖 cot 2 𝜃
42
The Kepler problem in action-angle variables (cont.)

By a change of variable to 𝑢, defined through

sin 𝑢 = cot 𝑖 cot 𝜃 , (10.163)

the integration can be performed trivially, and the expression for 𝑤1 reduces to

2𝜋𝑤1 = 𝜙 − 𝑢. (10.164)

The Figure 10.7 shows clearly that the

difference between 𝜙 and 𝑢 must be Ω, so that

2𝜋𝑤1 = Ω. (10.165)

43
The Kepler problem in action-angle variables (cont.)

In a similar fashion, we can identify the physical nature of the constant 𝑤2 .

Of the integrals making up 𝑊, Eq. (10.160), the two over 𝜃 and 𝑟 contain hand
are therefore involved in finding 𝑤2 .

By suitable choice of the arbitrary lower limit of integration, it can thus be

found that 2𝜋𝑤2 is the difference between two angles in the orbital plane, one of which
is the angle of the radius vector relative to the line of nodes and the other is the same
angle but relative to the line of the periapsis.

In other words, 2𝜋𝑤2 is the argument of the perihelion:

2𝜋𝑤2 = 𝜔. (10.166)

44